Some things just take time
lucumr.pocoo.org
lucumr.pocoo.org
I’ve been noticing that this simple reality explains almost all of both the good and the bad that I hear about LLM-based coding tools. Using AI for research or to spin up a quick demo or prototype is using it to help plot a course. A lot of the multi-stage agentic workflows also come down to creating guard rails before doing the main implementation so the AI can’t get too far off track. Most of the success stories I hear seem to be in these areas so far. Meanwhile, probably the most common criticism I see is that an AI that is simply given a prompt to implement some new feature or bug fix for an existing system often misunderstands or makes bad assumptions and ends up repeatedly running into dead ends. It moves fast but without knowing which direction to move in.
It also moves fast with a tendency to pick the wrong direction (according to the goal of the prompter) at every decision point (known or unknown).
This is a real problem when the "direction" == "good feedback" from a customer standpoint.
Before we had a product person for every ~20 people generating code and now we're all product people, the machines are writing the code (not all of it, but enough of it that I will -1 a ~4000 line PR and ask someone to start over, instead of digging out of the hole in the same PR).
Feedback takes time on the system by real users to come back to the product team.
You need a PID like smoothing curve over your feature changes.
Like you said, Speed isn't velocity.
Specifically if you have a decent experiment framework to keep this disclosure progressive in the customer base, going the wrong direction isn't a huge penalty as it used to be.
I liked the PostHog newsletter about the "Hidden dangers of shipping fast", I can't find a good direct link to it.
https://newsletter.posthog.com/p/the-hidden-danger-of-shippi...
As dev team we were able to crank the speed even more and silly product people thought they are doing something good by demanding even more from us. But that was one of the instances where users were helpful :).
People use dozens of apps every day to do their work. Just think about how are you going to make time to give feedback to each of each.
That's pretty much solved by the size of the audiences. You won't give feedback on 12 apps, but 11 other people will probably do so on 11 different apps.
Of course, the issue with my domain is that there's plenty of feedback, and product owners just dismiss it. Burn down your entire portfolio to get that boosted shareholder value for the next earnings report.
Well by asking repeatedly of course but you just piss people off.
Have you ever given feedback to Atlassian, Google, Microsoft?
This tayloristic idea (which has now reincarnated in "design thinking") that you can observe someone doing a job and then decide better than them what they need is ridiculous and should die.
Good products are built by the people who use the thing themselves. Doesn't mean though that choosing good features (product design and engineering) isn't a skill in itself.
sure 'everyone' drives so you can be a domain expert in cars. However not everyone can be an astronaught - rockets are complex enough to need more people than astronaughts and so most people designing spaceships will never have the opportunity to use one.
If you're bottle-necked by waiting for the users of your product to give a feedback, you clearly need to spend more time learning how to be a user yourself. Or hire people with some domain skill who can also code.
Yes you're right not anyone can be a domain expert. But anyone in the company needs to at least try to use the product as much as possible.
I worked in companies where even the CEO had never used the product but was telling us what to implement.
I suppose there is an argument that if you are building the wrong thing, build it fast so that you can find out more quickly that you built the wrong thing, allowing you to iterate more quickly.
Way too often that is used as an excuse for various forms of laziness; to not think about the things you can already know. And that lack of thinking repeats in an endless cycle when, after your trial and error, you don't use what you learned because "let's look forward not backward", "let's fail fast and often" and similar platitudes.
Catchy slogans and heartfelt desires are great but you gotta put the brains in it too.
I doubt GP is suggesting ‘go ahead and be negligent to feedback and guardrails that let you course correct early.’
Plugging the Cynefin framework as a useful technique for practitioners here. It doesn’t have to be hard to choose whether or not rigorous planning is appropriate for the task at hand, versus probe-test-backtrack with tight iteration loops.
A lot of people are so enamored by speed, they are not even taking the time to carefully consider the full picture of what they are building. Take the HN frontpage story on OpenCode: IIRC, a maintainer admitted they keep adding many shallow features that are brittle.
Speed cannot replace product vision and discipline.
This won't really stop until investors start judging on quality and not quantity. But a lot of those are thinking in finances, and the thought of removing their biggest cost center is too tempting to not go all in on. So they want to hear "we made this super fast with 2-3 people!" instead of "we optimized and scaled this up to handle 400% more workload with double the performance".
Discovery is great and all but if what you discover is that you didn't aim well to begin with that's not all that useful.
Before AI, I worked at a B2B open source startup, and our users were perpetually annoyed by how often we asked them to upgrade and were never on the latest version.
And frankly, they were in point.
Especially in the B2B context stability is massively underrated by the product.
There is very little I hate more then starting my work week on a Monday morning and find out someone changed the tools I'm using for daily business again
Even if it's objectively minor like apples last pivot to the windows vista design... It just annoys me.
But I'm not the person paying the bills for the tools I'm using at work, and the person that is almost never actually uses the tools themselves and hence shiny redesigns and pointless features galore
In the last year or so I've been able to prototype it and accelerate the development quite significantly using Claude and pals, and now it is very close to a finished product. One one hand there's no doubt in my mind that the LLM tools can make this sort of thing faster and let you churn through ideas until you find the right ones, but on the other hand, if I hadn't had that slow burn of mostly just thinking about it conceptually for 10 years, I would have ended up vibe coding a much worse product.
Speed actually just wins, because we are usually constrained by time.
Working or useful software? AI hasn't produced any at all since 2023.
Sorry, but I don’t understand what you mean here. What do we win by being faster at producing the wrong things?
1) a lot of shallow, orthogonal directions is better than 1 deep, careful approach
2) There's no social aspect to churning out a bunch of slop that will affect the perception of potential "right things" later. My domain can be particularly grudgeful in this regard.
2) I read that part twice and could not figure out what it is you are trying to say.
A proper capitalist system will tend toward the right direction as directed by the market yea? All of this neuroticism about AI doesn't matter.
This reminded me of the idea that civilization is already a misaligned superintelligence, and that technology (incl. AI) just moves it faster in the wrong direction.
That's basically the problem of supermorality. If you're an actually benevolent AI, do you do what civilization tells you? Or do you do what is good? What happens if you disagree?
Thus far, AI has been good... For venture capitalists. Jury's out if it's good for humanity and civilization at large. There have been a lot of benevolent usages of AI thus far, but also a lot of bad.
As for those who disagree with the "benevolent AI," I think they just get sent to the gallows (either metaphorically or literally)
Exactly this. Velocity is a vector. It has magnitude (aka speed) and direction.
Our industry has chased magnitude over all else for so long. Now we can put nitro in everyone's car and we get to where we wish to go very fast. Suddenly bad direction-setting is getting feedback where there used to be friction and natural time to steer.
My greatest hope is that a ton of bad leaders and middle managers end up finally getting exposed due to the advent of AI. (Will I be disappointed? Almost certainly yes.)
The number of times I've seen them get the wrong end of the stick in their COT is ridiculous.
Even when I tell them to only implement after my explicit approval they ignore this after 2 or 3 followups and then it's back to them going down blind alleys.
If you need to spend actual humans searching every 10 degrees (or more if they need to go far), the cost might not be worth it. What if they don't come back? Every human scouting is not foraging for the group or protecting them from whatever is in the jungle,
Now if you have a near infinite capacity of autonomous drones you can send one every degree from your position and get an approximate estimate whether that direction is safe or not.
It's the same with programming, you can do a TON of exploratory crap with near zero extra cost. Just to see if that's the right direction or not. Basically lean in to the "fast but inaccurate" aspect of AI.
Like I had a hare-brained idea of a local web-based file manager / media browser that would use YOLOv8 and CLIP to tag and classify images and videos as you browse. Took me about 30 minutes of active time over a few evenings while doing other stuff to conclude that, yes, it IS doable. Now I can focus my actual brain on it.
But I am not just dispatching to agents. I work interactively with a chat interface, and sometimes, I will just bin a whole hour's worth of back-and-forth, because we're not getting anywhere (in fact, I did exactly that, about 30 minutes ago).
But that hour is peanuts, compared to the ten hours that I would have spent, trying to figure it out on my own. With an LLM (and git), I can "run something up the flagpole, and see who salutes." I can afford to experiment with very large code bases, and toss out a whole bunch of stuff, if need be.
That said, I know damn well, that quite a few folks here, would sneer at my methodology, as "awkward, stodgy, and slow." Nevertheless, I am pretty chuffed with the results. Yeah, it's slower than some folks would do it, but the Quality is really high, and I'm happy with the results.
My favorite thing to do, is (for example) toss all 5 of my SDK files into the LLM, paste in the JSON server interaction, describe the bug, and ask it to help me figure it out.
Nine times out of ten, it finds the bug quickly. The real bug. I am not always happy with the proposed solutions, but finding the root cause is always the time-consuming part.
One more thing I try is give same prompt once in a while to ChatGPT/Gemini/Grok. Than take the 2 out of 3 ideas forward.
All leading AI seems to have some blind spot. Like some kind of intricsic character. 1 will completely overlook some particular bug while finding other excellent bugs and edge cases.
Getting the code through all 3 before commiting has shown excellent results to me.
>toss all 5 of my SDK files into the LLM, paste in the JSON server interaction, describe the bug, and ask it to help me figure it out.
I wonder if this "quality" code wouldn't have that many bugs to dive into if it was more carefully considered and produced up front?
This harkens back to a study in 2024 were senior devs were actually less productive with LLMS but they felt more productive, even after being told they were less productive.
I'm certain this is true for me. The only thing LLMs do get me is the ability to make forward progress on tasks while I'm in meetings, which is a net positive at least.
I’m actually a terrible programmer. This is just how I compensate.
Of course, you are more than welcome to visit my portfolio (easily found from my HN profile), and make fun of it.
Its just more that we should understand that there's no free lunch.even pre-Ai, codegen has the offset of creating nigh unmaintainable code, and making proper iteration (changing the codegen itself) take longer. But ideally, you gain long term speed at the cost of massive tech debt.
This ultimately operates as faster codegen, wrapped in a black box. If shipping is all that matters, then it is what it is. But we should understand the sacrifices for that.
I am not yet comfortable with just asking an agent to "write me a Facebook." That seems to be what all the kids are into, these days, but I am not convinced the tech has reached these lofty expectations.
I just use it as I would, a coding partner, who knows a lot, is a bit obsessive, and does things quickly. I've worked with some aspies, in my day, that actually come very close. Working with them, has helped me to work with an LLM, better.
I have been around, long enough, to have seen several "watershed" developments in software technology. Each one, was accompanied by wailing, rending of clothes, and gnashing of teeth; yet, here we are. We have not just survived. We have thrived.
For an experienced coder, writing code is never the problem. Some tasks are more raw writing effort then others, sure. The problem is understanding the problem, the task, the constraints and how to express them correctly, in a way that suits your environment (i.e. company, ecosystem, platform).
My brain is the actual problem, how fast I can digest information and make sense of it. For me, this is the process that always consumes the most time.
Last week I've wasted a full day, just because I didn't had a decent overview of what I was working on. It was heavily LLM chatbot assisted, at no point in time the LLM nudged me into the right direction, mainly because I was confidently asking the wrong questions.
No AI will help you if you insist being dumb, but it will accelerate you in that regard. That at least works for me.
I've been using AI to help me write it and I've come to a couple conclusions:
- AI can make working PoCs incredibly quickly
- It can even help me think of story lines, decision paths etc
- Given that, there is still a TON of decisions to be made e.g. what artwork to use, what makes sense from a story perspective
- Playtesting alone + iterating still occurs at human speed b/c if humans are the intended audience, getting their opinions takes human time, not computer time
I've started using this example more and more as it highlights that, yes, AI can save huge amounts of time. However, as we learned from the Theory of Constraints, there is always another bottleneck somewhere that will slow things down.
Coming up with a genuinely interesting gameplay loop with increasing difficulty levels and progressively revealed gameplay mechanics is a fascinating and extremely difficult challenge, no matter how much AI you throw at the problem.
If LLMs (or other "AI" or even AI tools) are able to exactly replicate the behavior of a program (game or otherwise) without access to its source code, that's technologically cool. However, that means it's possible to cheaply replicate immense amounts of human work in a way the law does not cover.
If you take a game and use LLMs to reimplement both its assets and code from scratch but players have the same movement, weapons do the same damage, have the same spread and projectile speed, and so on, then the "new" game is not really new, it's based on other people's work. And nobody should be allowed to profit from other people's work without their consent and without compensating them.
Obviously, work is hard to quantify but that doesn't mean we should give up.
2) We can also look at it from a more utilitarian perspective. When something starts as closed source, people who made it got paid already and the owners (who often did not perform any useful work except putting in money) keep making money from then on. Reimplementing it as open source does not harm the original devs but allows more people to access it and it also often leads to a much more open and pro-social implementation without dark patterns. And the paid version often still has an advantage due to existing awareness, marketing and network effects.
OTOH when something starts as free/open under conditions such as anyone building on top of it has to release under the same conditions, then a company taking that work is violating explicitly stated wishes, is making money which doesn't reach the original devs and does not promote the original work. And it also has the aforementioned advantages. When the closed version eclipses the open one, the owners are free to add dark patterns and otherwise exploit their position further.
This way open work is a global social good, closed work is only good for those who own it.
---
I prefer argument 1 because it doesn't require the presence of exploitative power structures.
Either way, we should recognize there are multiple dimensions to compensation - here recognition and money. And work should be rewarded along both axes transitively.
„I was able to vibecode those 5 apps I always wanted but never had time to code them myself … it is so different now because — I don’t have time to use them”.
However an app needs to be a good enough ui that it does something for you and often they don't. Paper todo lists still beat all the todo apps people try to make because the ui works for paper. Paper has large limits and so it seems there must be an app out there that is better - but so far everyone has failed to create it.
Idea behind the joke is that those apps were not needed at all in the first place, it is only that person from the joke had grand idea lurking in their head that those will be super useful. If those apps would be needed a person would take time to build them. Now building useless apps costs much less.
Better UI doesn't matter if something is purely a "wannabe application" in someones head.
Ergo, if you don't have time to build out some idea - that is a clear indication such idea is not good or not so useful - even if you still have an urge to build it, subconsciously you already know.
Really like the article I think it is awesome, and I strongly believe AI for coding will stay, but I also beleive that we need to still have a strong understanding of why we are building things and what they look like.
Lost me in paragraph three. We pay for those things because they're recognizable status symbols, not because they took a long time to make. It took my grandmother a long time to knit the sweater I'm wearing, but its market value is probably close to zero.
The fact that those items took a long time to make is part of what makes them status symbols though, because if you pay a lot of money for something that took no time to make at all (see most NFTs) you look like an idiot to a lot of people.
This sort of thing was done at a time when everybody did it, and now that it's not done, nobody does it
No kid ever said "did you see the sweater that Timmy's grandma knitted for him? That kid is so cool! "
Mostly because they all had grams sweaters as well.
I don't know what term you were looking for, but a handmade present for someone dear is about the furthest thing from a "status symbol" that I can think of:
- it can't be bought
- it can't be transferred without losing almost all value (ie: it's only valuable to you, or at most your family, eBay doesn't want it)
- it provides no improvement whatsoever in one's social standing
I'm also completely unimpressed by someone wearing a Rolex though, so different mileage for different people.
Understanding words does not require being impressed by anything, nor caring about the opinion of kids.
If people don't consider that someone with more money is of a higher status then symbols of that wealth aren't meaningful.
I think a lot of people have an ingrained belief that "more money == more status"
The old rich doesn't give a shit about Rolex watches beyond noticing the newb rich using them to tell on themselves.
I can't connect it at all to your listed points. An Olympic medal is about obvious a status symbol as I can imagine but it can't (meaningfully) be bought or transferred.
The status signified with a knit sweater is membership (and good standing!) in a caring family with elders not yet fully subsumed into their phones.
People, acquaintances and strangers alike, frequently comment on the knit socks I often wear, ask after who made them, and all of a sudden we're on "how's your mom" terms.
https://www.ebay.com/b/Olympic-Medal/27291/bn_55191416?_sop=...
> People, acquaintances and strangers alike, frequently comment on the knit socks I often wear,
Ok, that explains pretty much everything about your line of thought.
Thanks.
> https://www.ebay.com/b/Olympic-Medal/27291/bn_55191416?_sop=...
Of course you can buy an Olympic medal. You can't buy the status conferred by the medal (of Olympic champion / nth runner up).
> Ok, that explains pretty much everything about your line of thought.
I don't understand this either. Are you insulting me?
Parents want to signal "this child is looked after and we have a lot of capacity". Clothes, lunches, a lot of things are quietly like this.
A good hand-made gift demonstrates the status of the giver and provides proof-of-work for their regard of the recipient.
Some are indeed via marketing, but any itema have intrinsic or at least, emotional value.
The sweater is with whatever value a single person values it as or would pay for it. Said another way, would you sell it to me for $10? 50? 100? If you said no to all three, it's worth at least $100.
You fill a jar with sand and there is no space for big rocks.
But if you fill the jar with big rocks, there is plenty of space for sand. Remove one of the rocks and the sand instantly fills that void.
Make sure you fit the rocks first.
You fill the bottle with half of the water, you put the fish in, you can fill in the other half. If you start with the first half, you will end up with more water.
Not sure, I used to be better at diagnosing this type of episode.
> Given a 3-liter container and a 5-liter container, both initially empty, and access to tap water, how can you measure exactly 4 liters of water without using any additional containers
I've offered and received some convoluted metaphors recently, love leaning hard into this one.
then you fill 3 liter bottle again, and pour the contents into the 5 liter bottle until the 5 liter one is full
empty the 5 liter bottle, and pour the 1 liter in the 3 liter bottle into the 5 liter bottle
fill the 3 liter bottle again and pour that into the 1 liter already in the 5 liter bottle to get 4 liters of water
That water overflow step is missing / implicit. But that's an observable event.
Yes, there are bad metaphors, and people who take metaphors too seriously. That you can conjure a bad metaphor with somewhat similar to semantics to some other metaphor does not mean that said metaphor is bad.
An example being the common attitude that [advanced tech] is just a math problem to be solved, and not a process that needs to play itself out in the real world, interacting with it and learning, then integrating those lessons over time.
Another way to put this is: experience is undervalued, and knowledge is overvalued. Probably because experience isn’t fungible and therefore cannot be quantified as easily by market systems.
1. Probably not his original idea, and now that I think about it this is kind of more Hegelian. I’m not familiar enough with Hegel to reference him though.
Resource to be exploited. That’s worse of course.
Agentic coding very much feels like a "video game" in the sense of you pull the lever and open the loot box and sometimes it's an epic +10 agility sword and sometimes its just grey vendor trash. Whether or not it generates "good" or even "usable" code fades to the background as the thrill of "I just asked for a UI to orchestrate micro services and BLAMMO there it was!" moves to the fore.
Imagine a world in which the promise of AI was that workers could keep their jobs, at the same compensation as before, but work fewer hours and days per week due to increased productivity.
What could you do with those extra hours and days? Sleep better. Exercise more. Prepare healthy meals. Spend more time with family and friends. The benefits to physical and mental well-being are priceless. Even if you happened to earn extra money for the same amount of work, your time can be infinitely more valuable than money.
Unfortunately, that's not this world. Which is why the "increased productivity" promise doesn't seem to benefit workers at all.
If you look at the technological utopias that people imagined 50, 60+ years ago, they involved lives of leisure. If you would have told them that advances in technology would not reduce our working hours at all, maybe they would have started smashing the machines back then. Now we're supposed to be happy with more "stuff", even if there's no more time to enjoy stuff.
At one point, it was this world[1]:
> Consider a typical working day in the medieval period. It stretched from dawn to dusk (sixteen hours in summer and eight in winter), but, as the Bishop Pilkington has noted, work was intermittent - called to a halt for breakfast, lunch, the customary afternoon nap, and dinner. Depending on time and place, there were also midmorning and midafternoon refreshment breaks. These rest periods were the traditional rights of laborers, which they enjoyed even during peak harvest times. During slack periods, which accounted for a large part of the year, adherence to regular working hours was not usual. According to Oxford Professor James E. Thorold Rogers[1], the medieval workday was not more than eight hours. The worker participating in the eight-hour movements of the late nineteenth century was "simply striving to recover what his ancestor worked by four or five centuries ago."
> The contrast between capitalist and precapitalist work patterns is most striking in respect to the working year. The medieval calendar was filled with holidays. Official -- that is, church -- holidays included not only long "vacations" at Christmas, Easter, and midsummer but also numerous saints' andrest days. These were spent both in sober churchgoing and in feasting, drinking and merrymaking. In addition to official celebrations, there were often weeks' worth of ales -- to mark important life events (bride ales or wake ales) as well as less momentous occasions (scot ale, lamb ale, and hock ale). All told, holiday leisure time in medieval England took up probably about one-third of the year. And the English were apparently working harder than their neighbors. The ancien règime in France is reported to have guaranteed fifty-two Sundays, ninety rest days, and thirty-eight holidays. In Spain, travelers noted that holidays totaled five months per year.[5]
> The peasant's free time extended beyond officially sanctioned holidays. There is considerable evidence of what economists call the backward-bending supply curve of labor -- the idea that when wages rise, workers supply less labor. During one period of unusually high wages (the late fourteenth century), many laborers refused to work "by the year or the half year or by any of the usual terms but only by the day." And they worked only as many days as were necessary to earn their customary income -- which in this case amounted to about 120 days a year, for a probable total of only 1,440 hours annually (this estimate assumes a 12-hour day because the days worked were probably during spring, summer and fall). A thirteenth-century estime finds that whole peasant families did not put in more than 150 days per year on their land. Manorial records from fourteenth-century England indicate an extremely short working year -- 175 days -- for servile laborers. Later evidence for farmer-miners, a group with control over their worktime, indicates they worked only 180 days a year.
[1] https://groups.csail.mit.edu/mac/users/rauch/worktime/hours_...
Think about the analogy of transaction speed of money transfer vs actual delivery of good. With AI, we would make all digital tasks instantaneous, but the physical world will hum along at its own speed unless we speed it up with dark factories and what not.
I think of it differently. Speed is great because it means you can change direction very easily, and being wrong isn't as costly. As long as you're tracking where you're going, if you end up in the wrong place, but you got there quickly and noticed it, you can quickly move in a different direction to get to the right place.
Sometimes we take time mostly because it's expensive to be wrong. If being wrong doesn't cost anything, going fast and being wrong a lot may actually be better as it lets you explore lots of options. For this strategy to work, however, you need good judgment to recognize when you've reached a wrong position.
Creating these wrong things is only cheaper with LLMs. Since developers now spend less time and effort to create that wrong thing, they don't feel the need validate or reflect on them so much.
The risk is not the tool itself, but the over-reliance on it and forgoing feedback loops that have made teams stronger, e.g. debugging, testing, and reasoning why something works a particular way.
Yes, you cannot build years of community and trust in a weekend. But sometimes it's totally sufficient to plant a seed, give it some small amounts of water and leave it on its own to grow. Go ask my father having to deal with a huge maple tree, that I’ve planted 30 years ago and never cared for it.
Open Source projects sometimes work like this. I've created a .NET library for Firebase Messaging in a weekend a few years ago… and it grew on its own with PRs flowing in. So if your weekend project generates enough interest and continues to grow a community without you, what’s the bad thing here? I don’t get it.
Sometimes a tree dies and an Open Source project wasn’t able to make it.
That said, I’ve just finished rewriting four libraries to fix long standing issues, that I haven’t been able to fix for the past 10 years.
It's been great to use Gemini as a sparring partner to fix the API surface of these libraries, that had been problematic for the past 10 years. I was so quick to validate and invalidate ideas.
Once being one of the biggest LLM haters I have to say, that I immensely enjoy it right now.
Can't really say the same for vibecoding. You still need to do a lot of work that's ultiamtely putting lipstick on a pig. Maybe someone talented can make it pretty, but it has a quality ceiling, and most won't get anywhere close to that; people will just see a pig with lipstick on it.
I feel this new world sucks. We have new technology that boosts the productivity of the individual engineer, and we could be doing MUCH better work, instead of just rushed slop to meet quotas.
I feel I'm just building my replacement, to bring the next level of profits to the c-suite. I just wish I wasn't burning out while doing so.
I don’t think it’s exclusive to startups or tech either, it seems more like a downstream consequence of the fact that there’s no real innovation anymore. Capitalism demands constant growth, and when there are real technological improvements you can achieve that growth through higher productivity. If there are none, you have to achieve that growth through other means like forcing employees to work longer or cutting costs. The alpha is all coming from squeezing the labor force right now.
This doesn't sound right to me. We are currently getting smacked upside the head by an enormous technological innovation. I believe that, even within the framework of capitalism, this problem has social and political roots. The "robber baron" period late 19th century America has strong similarities to what we are seeing today, and technological stagnation was not the cause.
And as of now, we are not having "technological innovation". We found a new jackhammer and are tearing up the entire house experimenting with it. Maybe when the "shiny new thing" effect wears off we'll get true innovation. But as of now people are just getting paid to show off jackhammers.
I started out building a full stack framework like Meteor framework (though I started before Meteor framework was created in 2012 and long before Next.js).
Then I ported it to to Node.js because I saw an advantage to having the same language on the frontend and backend.
Then I noticed that developers like to mix and match different libraries/modules and this was a necessity. The whole idea of a cohesive full stack framework didn't make sense for most software. So I extracted the most essential part of it that people liked and this became SocketCluster. It got a fair amount of traction in the early days.
At the time, some people might have thought SocketCluster was trying to be a more scalable copycat of Socket.io but actually I had been working on it for several years by that point. I just made the API similar when I extracted it for better compatibility with Socket.io but it had some additional features.
A few years ago, I ended up building a serverless low-code/no-code CRUD platform which removes the need for a custom backend and it can be used with LLMs directly (you can give them the API key to access the control panel). It can define the whole data schema for you. I've built some complex apps with it to fully prove the concept with advanced search functionality (including indexing with a million records).
I've made some technical decisions which will look insane to most developers but are crucial and based on 15 years of experience, carefully evaluating tradeoffs and actual testing with complex applications. For example my platform only has 3 data types. String, Number and Boolean. The string type supports some additional constraints to allow it to be used to store any kind of data like lists, binary files (as base64)... Having just 3 types greatly simplifies spam prevention and schema validation. Makes it much easier for the user (or LLM) to reason about and produce a working, stable, bug-free solution.
That said I've been struggling to sell it because there are some popular well funded solutions on the market which look superficially similar or better. Of course they can't handle all the scenarios, they're more complex, less secure, don't scale, require far more LLM tokens, lead to constant regressions when used with AI. It's just impossible to communicate those benefits to people because they will value a one-shotted pretty UI over all these other aspects.
You can check out https://saasufy.com/ if interested.
Saasufy itself isn't open source. I'm planning to sell licenses of the code (a limited number of them to make it scarce). SocketCluster is a core component of Saasufy. The goal did evolve slightly; originally, it was to make it easier to build full stack applications. Now it actually lets you build entire full stack apps without code. That bigger goal has been achieved. I have some videos linked from the Docs page showing how it works.
But yes, I'm a bit paranoid about my situation. I do feel like my work is suppressed by algorithms. Things feel very different for me now than they did before in terms of finding users. It's really hard to find people to try my work. Difficult even to convince them to watch a 10 minute video. Though I guess many people are in the same boat right?
It's not just for chat. Anytime people update data collaboratively, for any data-driven software, you run into issues of people overwriting each other's changes... Sadly users got really used to poor experiences like having to 'refresh the page', or 'locking down resources' during edits to avoid concurrent editing altogether.
It's been kind of disturbing for me to realize how few people care about data integrity in collaborative scenarios. It's a kind of hidden issue which can cause major problems but users tend to tolerate. I've had this experience first-hand with colleagues using various other platforms where someone else had the same resource open for a while in the browser, we updated different fields and my field was overwritten with its old value but we didn't realise for some time until it caused a bug with one of the workflows. Then it was like "I'm sure I updated that value."
SC's pub/sub channels are part of the secret sauce to make this work. SocketCluster makes some unique tradeoffs with its pub/sub mechanism; the channels are extremely cheap to create and are cleaned up automatically (and cheaply)... And at any scale with sharding. This was an intentional design decision which is different from message queues like Kafka and RabbitMQ which prioritise throughput and are typically used in the backend as opposed to being end-to-end.
I do not think that it's your competitors' pretty one-shotted UIs that are losing you your leads. FWIW your website:
* Does not explain the benefits nor the feature-set of what your platform offers
* Has a Docs page which seems to be a link to 6 sets of reference APIs (or similar) in disparate Github pages - that's worse than most open source products.
* Uses cute animated characters which does not give off the "we are a serious enterprise and you can rely on us" vibe
* Has a novel pricing scheme ("implements a techno-feudalist pricing model") that requires a 4 paragraph explanation and somehow tries to frame introducing additional uncertainty for potential buyers as a feature (it's a good thing that _my_ pricing tiers could be adjusted by other customers' votes!?)
* Additionally prominently features cryptocurrency in the topnav, which is for many people a yellow or red flag (regardless of what seems to be your good intentions behind it)
Doesn't have any demo apps I can click around in to smoke test the platform's functionality, let alone fiddle with the backend
* Has almost zero information on basic business-y compliance-y things - no info on security, availability, SSO support, etc let alone more hardcore things like compliance standards your platform meets.Also I'll try to explain like "Estimated to be X times faster to develop and requires 70% fewer tokens than alternative stacks and easier to configure security."
That's actually realistic based on some recent analysis I did. I didn't have this info before.
What's funny, is that I had many other projects attempted over the years, but many rise and die quickly, yet the one that lasted the longest is also the one that is likely to last longest from now.
[0]: https://uxwizz.com
Consider the idea of trying to determine how quickly an unknown number of timers will go ping, It could be 10,000 timers that go ping when finished or 1,000,000 timers that go ping when finished. I don't know when they are going to go ping, just that they all the timers are running at different speeds spread over some distribution.
After one time period, 5,000 pings have been detected. Should you conclude that timers are pinging fairly quickly?
You cannot tell the overall duration of timers if you don't know the number of timers there are out there. Your only data that the timer exists is the ping, consequently you cannot tell if a small population is at high speed or a large population is at a moderate speed. In both cases the data you receive are the fastest of the population.
In other words we haven't yet seen what the 10 year project made using these tools is like (or even if it exists/will exist), because they haven't been around for 10 years.
Btw the earendil.com website is gorgeous.
It really is but it's come at the cost of actually being useful. It has a vague 'about' modal and that's your lot, which is confusing since they're encouraging people to join. I'm just not sure anybody's going to know what.
It's a surprisingly good filter and gets curious people to send mail in :)
Undoubtedly a lot of that comes down to production cost and safety. A plane is far more likely to kill people and it costs a shitload more to produce then an app (though plenty of software is mission critical). But now in software we can move quick enough up front that if we don't start applying some discipline it's going to bite us in the ass in the long run.
I do wonder if productivity with AI coding has really gone up, or if it just gives the illusion of that, and we take on more projects and burn ourselves out?
ps: it's strange that YouTubers are talking about the same thing. People in different dev circles. Agentic feels like doom ide scroll.
It definitely hasn't for me. I spent about an hour today trying to use AI to write something fairly simple and I'm still no further forward.
I don't understand what problem AI is supposed to solve in software development.
When Russians invaded Germany during WWII, some of them (who had never seen a toilet) thought that toilets were advanced potato washing machines, and were rightfully pissed when their potatoes were flushed away and didn't come back.
Sounds like you're feeling a similar frustration with your problem.
Why is AI supposed to be good?
I ended up having to type hundreds of lines of description to get thousands of lines of code that doesn't actually work, when the one I wrote myself is about two dozen lines of code and works perfectly.
It just seems such a slow and inefficient way to work.
I've also been underwhelmed with its ability to iterate, as it tends to pile on hacks. So another useful question is "did you try having it write again with what you/it learned?"
In my case I'd have to say "Don't know, whatever VS Code's bot uses", and "no idea what those are or why I have to care".
The reason I ask about what model is I initially dismissed AI generated code because I was not impressed with the models I was trying. I decided if I was going to evaluate it fairly though, I would need to try a paid product. I ended up using Claude Sonnet 4.5, which is much better than the quick-n-cheap models. I still don't use Claude for large stuff, but it's pretty good at one-off scripts and providing advice. Chances are VS Code is using a crappy model by default.
> no idea what those are or why I have to care
For the difference between chat mode and agent mode, chat mode is the online interface where you can ask it questions, but you have to copy the code back and forth. Agent mode is where it's running an interface layer on your computer, so the LLM can view files, run commands, save files, etc. I use Claude in agent mode via Claude Code, though I still check and approve every command it runs. It also won't change any files without your permission by default.
AGENTS.md and CLAUDE.md are pretty much a file that the LLM agent reads every time it starts up. It's where you put your style guide in, and also where you have suggestions to correct things it consistently messes up on. It's not as important at the beginning, but it's helpful for me to have it be consistent about its style (well, as consistent as I can get it). Here's an example from a project I'm currently working on: https://github.com/smj-edison/zicl/blob/main/CLAUDE.md
I know there's lots of other things you can do, like create custom tools, things to run every time, subagents, plan mode, etc. I haven't ever really tried using them, because chances are a lot of them will be obsolete/not useful, and I'd rather get stuff done.
I'm still not convinced they speed up most tasks, but it's been really useful to have it track down memory leaks and silly bugs.
Okay. Get me a job and I'll pay for any model of your choosing. Until then, finances are very slim.
Heh, I'm a college student, so I can't help with that...
You could also try Gemini 3 pro with Gemini's CLI which is free, though it's not as good at using tools. But, it sounds like you're not interested, which is fine!
Just please don't continue to argue with finer points if you're not interested. I've done my best to engage with your points, but I get the sense that it doesn't matter what I say.
I am curious though, why do you feel so strongly about LLM products?
But sure:
>why do you feel so strongly about LLM products?
Personally, I work in games. So pretty much everything in the discourse of LLMs and Gen AI has been amplified 5x for me. The layoffs, the gamers' reaction to stuff utilizing AI, the impact on hardware prices, the politics, etc.
Theres a war of consumers and executives, and I'm trapped in the middle taking heat from both. It's tiring and it's clear who to blame for all of this. I want all of this to pop so the true innovation can rise out, instead of all the gold rush going on right now.
Also,game code is very performance sensitive. It's not like a website or app where I can just "add 5 seconds to a load time" unless I'm working on a simple 2D game, nor throw more hardware to improve performance. Even if LLMs could code up the game, I'd spend more time optimizing what it makes than it saved. It simply doesn't help for the kind of software I work with.
> Personally, I work in games. So pretty much everything in the discourse of LLMs and Gen AI has been amplified 5x for me. The layoffs, the gamers' reaction to stuff utilizing AI, the impact on hardware prices, the politics, etc.
> Theres a war of consumers and executives, and I'm trapped in the middle taking heat from both. It's tiring and it's clear who to blame for all of this. I want all of this to pop so the true innovation can rise out, instead of all the gold rush going on right now.
That makes a lot of sense. I've been pretty fed up with the hyperbole and sliminess, and I can't imagine how difficult it is to be squeezed between angry gamers and naive and dense executives.
When you say "true innovation", is that in terms of non-AI innovation, or non-slimy AI innovation? I guess I personally still believe that LLMs are useful, but only as another tool amongst many others.
I'm also a big believer in human centered UX design, and it's kinda sad that the dominant experience is all textual.
> Also,game code is very performance sensitive
It does seem like game programming is the last bastion of performance, at least in terms of normal hardware, since the game has to go to the consumer's hardware. The "silver bullet" mentality drives me a little crazy because it clearly doesn't work in all situations.
Anyways, I don't know if this response really has a point, but I wanted to at least acknowledge your experience.
A bit of both. Similar to other tech investment, all the gaming centric accelerators are looking for is AI pitches. Makes me wonder what innovations thr past few years have been overlooked in lieu of the Ai Gold Rush.
But I can see the long term (likely 5+ years out) potebtial of Ai as well. Once we stop using it as a means to steal from and remove artists, I can see all kinds of tedious problems with assets that Ai can accelerate. Generative fill is a glimpse of a genuinely useful tool that helps artists instead of pretending to be an artist itself.
Can it eventually write performant code? Maybe. The other big issue is that 1) a lot of code isn't online to train on and 2) a lot of that code is still a mess to process, with little standards to follow. Maybe it can help with graphics code (which is much more structured) in the near future.
From speaking to my friends in the industry, it seems like uptake for code is happening slowly, but unevenly, and the results are largely dependent on the level of documentation, which is often lacking. (I know of a few people using AI for (high-quality!) work on Godot, and their AIs struggle with many of the implicit conventions present in the codebase.)
With that being said, I would say that LLMs have generally been quite the boon for the (limited) gameplay work that I have done of recent. Because the cost of generation is so cheap [0], it is trivial to try something out, experiment with variations, and then polish it up or discard it entirely.
This also applies to performance work: if it's a metric that the AI can see and autonomously work on, it can be optimised. This is, of course, not always possible - it's hard to tell your AI to optimise arbitrary content - but it's often more possible than not, especially if you get creative. (Asking it to extract a particularly hot loop out from the code it resides within, and then optimising that, for example: entirely feasible.)
I think there are still growing pains, but I'm confident that LLMs will rock the world of gamedev, just like they're doing to other more well-attested fields of programming.
[0]: https://simonwillison.net/guides/agentic-engineering-pattern...
Yeah, that sums up a lot of my thoughts with AI c. 2026.
I do take some schedenfreude knowing that AI training also struggles with the utter lack of documentation here. That may be a win in and of itself if this paradigm forces the games industry to properly care for tech writing.
>Because the cost of generation is so cheap [0], it is trivial to try something out, experiment with variations, and then polish it up or discard it entirely.
Well, that's another thing I'm less confident about. The cost is low, for now. But we also know these companies are in loss leader mode. It'll probably always be cheap for a company to afford agents, but I fear reliance on these giant server models will quickly price out ICs and smaller work environments.
That might be something China beats us too. They seem to be focusing on optimizing models that works on local machines out of necessity, as opposed to running tens of billions of dollars of compute. My other big bias is wanting to properly own as much of my pipeline as possible (to the point where my eventual indie journey is planning around OS tools and engines, despite my experience in both Unity and UE), and current incentives for these companies don't want that.
Shouldn’t have to baby step through the basics when the author is clearly not interested in learning himself
I'd rather assume good faith, because when I first started using LLMs I was incredibly confused what was going on, and all the tutorials were grating on me because the people making the tutorials were clearly overhyping it.
It was precisely the measured and detailed HN comments that I read that convinced me to finally try out Claude, so I do my best to pay it forward :)
But the guy is being adversarial and antagonistic. Its a 2 way street, sometimes you have to call people out on their BS because I'm not seeing someone argue in good faith, but rather pretending some superior knowledge because hes working on a esoteric protocol like people here don't know how packet headers work
> sometimes you have to call people out on their BS
That's true, but I think that it's often much later than what some people would consider enough. Someone can be bitter, and still have good points. It's very dangerous to preemptively dismiss points, because it means that I won't listen to anyone who disagrees with me. I'm willing to put in the work to interpret someone's response in a productive light because there's often something to find.
There's a framework that I work within when I'm in a discussion. There's three elements: arguments, values, and assumptions. An argument is the face value statements. But those statements come from the values and assumptions of the person.
Values are what people consider most important. In most cases, our values are the same, which is good!
The biggest difference is assumptions. For example, one assumption I have is that free markets are the best method we have to lift individuals out of poverty. This colors how I talk about AI. Another person might assume that free markets have failed, and we need to use a different approach. This colors how they would view AI. So we'll completely talk past each other when arguing AI, because it's more of a proxy war of our assumptions.
Trying to trace back the quality of the model to the "skills" of the person sounds extremely manipulative.
Okay. Whip up your favorite model and report back to us with your prompts. I'm pretty anti-AI, but you're going to attract more bees with honey than smoke.
The problem is that I want something that listens on a TCP connection for GD92 packets, and when they arrive send appropriate handshaking to the other end and parse them into Go structs that can be stuffed into a channel to be dealt with elsewhere.
And, of course, something to encode them and send them again.
How would I do that with whatever AI you choose?
I'm pretty certain you can't solve this with AI because there is literally no published example of code to do it that it can copy from.
No idea what you’re talking about but if it has a spec then it doesn’t matter if it’s trained on it. Break the problem down into small enough chunks. Give it examples of expected input and output then any llm can reason about it. Use a planning mode and keep the context small and focused on each segment of the process.
You’re describing a basic tcp exchange, learn more about the domain and how packets are structured and the problem will become easier by itself. Llms struggle with large code bases which pollute the context not straightforward apps like this
Okay, so you're running into the same problem that LLMs are.
> Break the problem down into small enough chunks. Give it examples of expected input and output then any llm can reason about it.
So I have to do lots of grunt work?
> You’re describing a basic tcp exchange, learn more about the domain and how packets are structured and the problem will become easier by itself
I've written dozens of things that deal with TCP. I already have a fully-working example of what I want. The idea was to test if I could recreate it using LLMs.
How is it supposed to work? How does it put in the code I already know I want?
I can't tell if you are a troll or not, but you can't complain that nobody understands your intentionally vague and obtuse way to describe the problem at hand to pretend you're superior.
https://www.publiccontractsscotland.gov.uk/NoticeDownload/Do...
You have to rename the file ending to PDF. It's probably the wrong spec, because I'm basing this research on literally four letters that could mean anything since there is zero context given here. I've also found some German documents about chemistry.
If your argument is that LLMs and humans are stupid because they don't know what a "GD92" is, then yeah maybe it's a you problem.
Go and throw the spec into openai codex inside limactl (get it from GitHub) and use zed (the editor) and a SSH remote project to get inside the VM, don't forget to enable KVM for performance. The free tier for openai is fine, but make sure to use codex 5.2.
First ask questions on what the binary encoding is based on. It's probably X.400, then once you've asked enough questions, tell it to implement it. You probably won't have to read the spec at all yourself.
Remember, I've already written something that does this. I'm trying to understand how and why an LLM would help.
Which part of the job is the LLM supposed to do?
I started the task 56 minutes ago with one prompt, and now I have an implementation I can show you. There's plenty to quibble about - the files splayed over the main directory are quite ugly, and there is no actual test data that we can use on the public internet - but these are all trivially resolvable issues.
I didn't do any additional research for this. I gave it the spec PDF, your instructions upthread, and told it to build a library. You can also consult the transcripts (linked in the README) to see that I have no tricks up my sleeve. I didn't need to decompose the task in any meaningful way: the only input I provided was on minor matters of taste.
Its not worth engaging a guy who is adversarial to learning how a tool works, just so he can maintain some air of superiority for his ego.
Here's the thing: we never had a remotely sane way to measure productivity of a software engineer for reasons that we all understand, and we don't have it now.
Even if we had it, it's not the sort of thing that management would even use: they decide how productive you are based on completely unrelated criteria, like willingness to work long hours and keeping your mouth shut when you disagree.
If you ask those types whether productivity has gone up with AI, they'll probably say something like "of course, we were able to let go a third of our programmers and nothing really seems to have changed"
"Productivity" became a poisoned word the moment that the suits realized what a useful weapon it was, and that it was impossible to challenge.
Not impossible to challenge. But most people don't have the legal funds to do so. Those that do tend to get a cushy severance bribe to stay quiet and they move on elsewhere.
That's also why it's a long process to "fire" someone but easy to "lay off" instead. layoffs are never about productivity (so it doesn't matter anyway), and the US is doing absolutely nothing to protect against it like most of the world.
It doesn’t matter how fast we can make our widgets and chatbots when what you need is to have a self sufficient workforce. We have outsourced everything material and valuable for society. Now we are left with industries of gambling, ad machines and pharmaceuticals with a government that is functionally bankrupt and politicians that have completely sold out
What's slower now are threats to production - even minor regulations take years or decades, and often appear only when workarounds have surfaced.
So what changed in the last 40+ years are the many tools for businesses to shape the conditions of their business -the downstream market, upstream suppliers, and regulatory support/constraints. This is extremely patient work over generations of players, sometimes by individuals, but usually by coalitions of mutual corporate self-interest, where even the largest players couldn't refuse to participate.
It's evolution.
I expect these tools will quickly let people to ramp up several orders of magnitude of more complexity and lines of code to any software project.
The your 100kloc JS electron app will become a 10m loc JS electron app running on a 500m loc browser runtime.
Repeat this across the stack for every software component and application and library. If you think things are bloated now just wait a few years and your notepad will be a 1m line behemoth with runtime performance of a glacier.
Anyway 2 areas I slightly disagree on.
Open source abandonware is fine. Sometimes people give up because they realize it is not a good idea. Or they get busy or sick.
And 10 years at a startup is great but that relies on it being a good startup. Entropy at companies means I have never made it to 10yrs even though I wanted to.
But anyhow, you can buy large-ish burlapped trees but they aren’t as healthy, often die, and nothing close to a 100+ yr old estate oak tree or a decades old rose garden. You just can’t make it faster, transplanting plants that old will kill them.
Most of the trees do just fine, and these nurseries will typically provide a warranty.
And forget about 20k. If you find someone willing to sell their tree you're looking at at least 10x that for the logistics of moving a 20 ton root system.
The desire to make life faster instead of better was identified as a neurosis of civilization in the 19th century.
Is it a bad sign that I am only skimming diagonally the AI summary of an article called "some things just take time" because it feels like I already know the idea?
This website does not resonate with the message I got earlier from the article. It does not give the impression of someone taking appropriate time to make it.
This has been on my mind for a long time.
I used to be all about frictionlessness. Speed! Convenience! Make it smoother!
Then I found... actually, paper based systems work way better for me. Digital systems just turn into big piles of bloat. It's too easy to add stuff. So they grow until they collapse under their own weight.
(Take a look at your contacts list. How many should still be in there? How many did you add for a one time thing and then keep forever? Should there be a temporary folder? Shouldn't it be the default? That's how it works in nature!)
Ended up using paper as a temporary improvisation, then realized it solved all the problems I had with digital systems.
Friction is good.
With communication, it used to cost money to communicate. Now it's free, and we now have a sea of noise where most messages are "adding negative value", because they steal your time and energy.
Same with the app stores. Do any search and you find an ocean of slop! The gems drown in the sea of slop.
Friction is good.
absolutely although i wonder how different 'trust' is in the culture of tomorrow? will it 'matter' as much, be as cherished, as earned over the fullness of time?
i suspect it is a pendulum - and we are back to oak trees at some point - but which way is the pendulum swinging right now?
That's a mantra I learned when getting into technology.
Asking questions about how things work, why it is a certain way, or why a shortcut was made often give you far better insights than anything else.
Slowing down and understandng is great. With AI this is even easier. But choose wisely, brains get full.
And time takes money
Enough money to pay one’s bills while one tends the growing tree
And if we have a society that ensures everyone is given the dignity of time
We also get a society that reaps what some create with that time
But if we have a society that only rewards pushing money back up the hierarchy
Then we all lose our time and our nest eggs to those who have the most.
Maybe the response is to slow down a bit—revisit what matters, and use it with some sense of proportion and coherence.
It's valued because it's more dense. It grew slowly. Now wood grows fast and it's less dense.
Like this article, that's fine for many things - you just need wood - but not always.
Mass production of engineered structural lumber.[1]
Things take the time they take.
Don't worry.
How many roads did St. Augustine follow
before he became St. Augustine?
Oh, I thought it was because they're a way to show off about being rich.
> We require age minimums for driving, voting, and drinking because we believe maturity only comes through lived experience.
Even if she could reach the pedals, my 4yo doesn't have the attention span to drive. This isn't a "lived experience" thing, it's a physical brain development thing. IIRC the are effects with learning math, where starting earlier had limited impact on being able to move to certain more advanced topics earlier; ie there's more going on than just hours of experience.
The standard age for voting is also the age for being a legal adult. There are sound logical reasons that these ages should match.
The standard drinking age is due to pressure by activists, and AIUI is lower in other countries.
Maybe for some. I think these examples were carefully chosen. Hermès are made in France, "Swiss watch" doesn't automatically mean Rolex, though in that case Rolex does own most of their manufacturing (though there is a whole world of carefully made watches out there that don't cost 10K). As for old properties... there is a huge range there, but unless you are living in a castle, most people, at least my city, are likely silently thinking: "I'm so sorry for them that they have to live in that old house."
Refactoring decent sized components are an order of magnitude easier than it was, but the more important signal is still, why are you refactoring? What changed in your world or your world-view that caused this?
Good things still take time, and you can't slop-AI code your way to a great system. You still need domain expertise (as the EXCELLENT short story from the other day explained, Warranty Void if Regenerated (https://nearzero.software/p/warranty-void-if-regenerated) ). The decrease in friction does definitely allow for more slop, but it also allows for more excellence. It just doesn't guarantee excellence.
You can't trust us with self-care. There's just too many shiny toys out there!
What AI allow us is to do those things we would not have been able to prioritize before. To "write" those extra tests, add that minor feature or to solve that decade old bug. Things that we would never been able to prioritize are we noe able to do. It's not perfect, it's sometimes sloppy, but at least its getting shit done. It does not matter if you solve 10% of your problem perfect if you never have time for the remaining 90.
I do miss the coding, _a lot_, but productivity is a drug and I will take it.
This is a bad start. Louis XIV at Versailles and Marly famously made while forests appear or disappear overnight, to the utter dismay of Saint-Simon, the memorialist, who thought this was an unacceptable waste of money and energy.
And this was before the industrial revolution. Today I'm sure many more miracles happen every day.
Vibe slop-ing at supersonic speeds and waiting years to grow aren't the only options, there's something in between where you have enough signal to keep going and enough speed to not waste years on the wrong thing.
I feel that today's VCs have completely disregarded the middle and are focused on getting as big as possible as fast as possible without regard to the effect it's having on the ecosystem.
no we don't want to miss genuine ways to speed things up to improve our productivity so we can do other or more things
They have spent the last decade building processes and guardrails for getting consistent average performance from people. But now, some talented people who worked at those companies are building their own new companies without the overhead and moving much, much more quickly.
I think what we assume is "vibe slop at inference speed" is not as simple as people make it out to be. From a perspective, I think generally it might be people trying to save jobs.
I'm seeing more slop come out of larger, older companies than the new ones (with experienced operators).
And the speed is somewhat scary. For smaller team it doesn't take as much effort to build deep, beautiful product anymore.
The bottleneck was never the ability for a engineer to code. It was the 16 layers between the customer and the programmer which has vanished in smaller companies and is forcing larger ones to produce slop.
I'm reading Against The Machine by Paul Kingsnorth, and now reading this blog piece is hard not to make connections with the points of the book: the usage of the tree as a counter-argument for the machine's automation credo exposed in the blog post very much aligns with I've read so far.
"I keep coming back to the trees. I’ve been maintaining Open Source projects for close to two decades now. The last startup I worked on, I spent 10 years at. That’s not because I’m particularly disciplined or virtuous. It’s because I or someone else, planted something, and then I kept showing up, and eventually the thing had roots that went deeper than my enthusiasm on any given day. That’s what time does! It turns some idea or plan into a commitment and a commitment into something that can shelter and grow other people."
Good luck _creating_ this by trying to predict the next best word in the sentence.
But no one wants to go out of their house.
Social connections. Trust. Facetime. All matter more than ever.
Want a moatable software business? Know your customers on a personal level. Have a personal relationship. Know the people that sign the contracts, know their kids names, where they vacationed last winter, their favorite local restaurant.
Get out of the house.
Not true, we do this because the 99% of the time it's true, however there are people who would be perfectly competent and responsible to drive without living to the age of 16-18. Same with voting, there are humans who have a deep understanding and intelligence about politics at a younger age than suffrage. Equally there are people who will be reckless drivers at 40 and vote on whim at 60.
We have these rules not because sophistication only comes through lived experience, we have them because it's strongly correlated and covers of most error cases.
To take this to AI, run the model enough times with a higher enough temperature, then perhaps it can solve your challenges with a high enough quality - just a thought.
The reason we need to wait is that it takes time for some things to mature.